[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAH9NwWcehP+9bekxMoxF3QdO_7j5zwNLuCmHNvd-JCCV_x85jA@mail.gmail.com>
Date: Mon, 4 Nov 2024 14:21:42 +0100
From: Christian Gmeiner <christian.gmeiner@...il.com>
To: Maíra Canal <mcanal@...lia.com>
Cc: Melissa Wen <mwen@...lia.com>, Maarten Lankhorst <maarten.lankhorst@...ux.intel.com>,
Maxime Ripard <mripard@...nel.org>, Thomas Zimmermann <tzimmermann@...e.de>,
David Airlie <airlied@...il.com>, Simona Vetter <simona@...ll.ch>, kernel-dev@...lia.com,
Christian Gmeiner <cgmeiner@...lia.com>, dri-devel@...ts.freedesktop.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] drm/v3d: Add DRM_IOCTL_V3D_PERFMON_SET_GLOBAL
Hi Maíra,
>
> Thanks for your patch! Just some small nits.
>
Thanks for your review.
> On 31/10/24 17:51, Christian Gmeiner wrote:
> > From: Christian Gmeiner <cgmeiner@...lia.com>
> >
> > Add a new ioctl, DRM_IOCTL_V3D_PERFMON_SET_GLOBAL, to allow
> > configuration of a global performance monitor (perfmon).
> > Use the global perfmon for all jobs to ensure consistent
> > performance tracking across submissions.
>
> I'd mention how this feature can be useful for user-space (for example,
> the implementation of a Perfetto datasource).
>
That's a great idea .. will be done in V3.
> >
> > Signed-off-by: Christian Gmeiner <cgmeiner@...lia.com>
> >
> > ---
> > Changes in v2:
> > - Reworked commit message.
> > - Removed num_perfmon counter for tracking perfmon allocations.
> > - Allowing allocation of perfmons when the global perfmon is active.
> > - Return -EAGAIN for submissions with a per job perfmon if the global perfmon is active.
> > ---
> > drivers/gpu/drm/v3d/v3d_drv.c | 1 +
> > drivers/gpu/drm/v3d/v3d_drv.h | 8 ++++++++
> > drivers/gpu/drm/v3d/v3d_perfmon.c | 34 +++++++++++++++++++++++++++++++
> > drivers/gpu/drm/v3d/v3d_sched.c | 14 ++++++++++---
> > drivers/gpu/drm/v3d/v3d_submit.c | 10 +++++++++
> > include/uapi/drm/v3d_drm.h | 15 ++++++++++++++
> > 6 files changed, 79 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
> > index d7ff1f5fa481..3c89f0daa5b8 100644
> > --- a/drivers/gpu/drm/v3d/v3d_drv.c
> > +++ b/drivers/gpu/drm/v3d/v3d_drv.c
> > @@ -214,6 +214,7 @@ static const struct drm_ioctl_desc v3d_drm_ioctls[] = {
> > DRM_IOCTL_DEF_DRV(V3D_PERFMON_GET_VALUES, v3d_perfmon_get_values_ioctl, DRM_RENDER_ALLOW),
> > DRM_IOCTL_DEF_DRV(V3D_SUBMIT_CPU, v3d_submit_cpu_ioctl, DRM_RENDER_ALLOW | DRM_AUTH),
> > DRM_IOCTL_DEF_DRV(V3D_PERFMON_GET_COUNTER, v3d_perfmon_get_counter_ioctl, DRM_RENDER_ALLOW),
> > + DRM_IOCTL_DEF_DRV(V3D_PERFMON_SET_GLOBAL, v3d_perfmon_set_global_ioctl, DRM_RENDER_ALLOW),
> > };
> >
> > static const struct drm_driver v3d_drm_driver = {
> > diff --git a/drivers/gpu/drm/v3d/v3d_drv.h b/drivers/gpu/drm/v3d/v3d_drv.h
> > index cf4b23369dc4..a0d920ec2b1d 100644
> > --- a/drivers/gpu/drm/v3d/v3d_drv.h
> > +++ b/drivers/gpu/drm/v3d/v3d_drv.h
> > @@ -179,6 +179,12 @@ struct v3d_dev {
> > u32 num_allocated;
> > u32 pages_allocated;
> > } bo_stats;
> > +
> > + /* To support a performance analysis tool in user space, we require
> > + * a single, globally configured performance monitor (perfmon) for
> > + * all jobs.
> > + */
> > + struct v3d_perfmon *global_perfmon;
> > };
> >
> > static inline struct v3d_dev *
> > @@ -584,6 +590,8 @@ int v3d_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
> > struct drm_file *file_priv);
> > int v3d_perfmon_get_counter_ioctl(struct drm_device *dev, void *data,
> > struct drm_file *file_priv);
> > +int v3d_perfmon_set_global_ioctl(struct drm_device *dev, void *data,
> > + struct drm_file *file_priv);
> >
> > /* v3d_sysfs.c */
> > int v3d_sysfs_init(struct device *dev);
> > diff --git a/drivers/gpu/drm/v3d/v3d_perfmon.c b/drivers/gpu/drm/v3d/v3d_perfmon.c
> > index 156be13ab2ef..bf42303c292b 100644
> > --- a/drivers/gpu/drm/v3d/v3d_perfmon.c
> > +++ b/drivers/gpu/drm/v3d/v3d_perfmon.c
> > @@ -312,6 +312,9 @@ static int v3d_perfmon_idr_del(int id, void *elem, void *data)
> > if (perfmon == v3d->active_perfmon)
> > v3d_perfmon_stop(v3d, perfmon, false);
> >
> > + /* If the global perfmon is being destroyed, set it to NULL */
> > + cmpxchg(&v3d->global_perfmon, perfmon, NULL);
> > +
> > v3d_perfmon_put(perfmon);
> >
> > return 0;
> > @@ -451,3 +454,34 @@ int v3d_perfmon_get_counter_ioctl(struct drm_device *dev, void *data,
> >
> > return 0;
> > }
>
> [...]
>
> > index 87fc5bb0a61e..709724fe28e6 100644
> > --- a/include/uapi/drm/v3d_drm.h
> > +++ b/include/uapi/drm/v3d_drm.h
> > @@ -43,6 +43,7 @@ extern "C" {
> > #define DRM_V3D_PERFMON_GET_VALUES 0x0a
> > #define DRM_V3D_SUBMIT_CPU 0x0b
> > #define DRM_V3D_PERFMON_GET_COUNTER 0x0c
> > +#define DRM_V3D_PERFMON_SET_GLOBAL 0x0d
> >
> > #define DRM_IOCTL_V3D_SUBMIT_CL DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_SUBMIT_CL, struct drm_v3d_submit_cl)
> > #define DRM_IOCTL_V3D_WAIT_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_WAIT_BO, struct drm_v3d_wait_bo)
> > @@ -61,6 +62,8 @@ extern "C" {
> > #define DRM_IOCTL_V3D_SUBMIT_CPU DRM_IOW(DRM_COMMAND_BASE + DRM_V3D_SUBMIT_CPU, struct drm_v3d_submit_cpu)
> > #define DRM_IOCTL_V3D_PERFMON_GET_COUNTER DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_PERFMON_GET_COUNTER, \
> > struct drm_v3d_perfmon_get_counter)
> > +#define DRM_IOCTL_V3D_PERFMON_SET_GLOBAL DRM_IOW(DRM_COMMAND_BASE + DRM_V3D_PERFMON_SET_GLOBAL, \
> > + struct drm_v3d_perfmon_set_global)
> >
> > #define DRM_V3D_SUBMIT_CL_FLUSH_CACHE 0x01
> > #define DRM_V3D_SUBMIT_EXTENSION 0x02
> > @@ -765,6 +768,18 @@ struct drm_v3d_perfmon_get_counter {
> > __u8 reserved[7];
> > };
> >
> > +#define DRM_V3D_PERFMON_CLEAR_GLOBAL 0x0001
> > +
> > +/*
>
> Please, use kernel-doc [1]. So, here is /**
>
> [1] https://docs.kernel.org/doc-guide/kernel-doc.html
>
> > + * struct drm_v3d_perfmon_set_global - ioctl to define a
>
> The colunm width is 80.
>
> > + * global performance counter that is used if a job has
>
> s/counter/monitor
>
> > + * not assigned one on its own.
>
> This description isn't really precise. When the global performance
> monitor is enabled, the job perfmon doesn't matter. Currently, we don't
> even allow a job perfmon to be submitted, but if we did, the global
> perfmon would be used even if the job had a perfmon.
>
I have rephrased this comment and should be more precise now.
--
greets
--
Christian Gmeiner, MSc
https://christian-gmeiner.info/privacypolicy
Powered by blists - more mailing lists